OpenAI and US AI chip unicorn Cerebras jointly announced a...A three-year cooperation agreementOpenAI will deploy a 750MW (megawatt) Cerebras wafer-scale system to create the world's largest high-speed AI inference platform. According to sources, the deal is worth over $100 billion (approximately NT$3200 billion).
Launched in 2026, focusing on the speed of "reasoning".
This deployment plan is expected to be implemented in phases starting in 2026 and completed by 2028. Cerebras' chip design logic is different from traditional GPUs. It is known for its "largeness" and boasts that it can pack 4 trillion transistors into a giant chip.
By integrating massive computing power, memory, and bandwidth onto a single chip, Cerebras eliminates the bottleneck that traditional hardware causes inference speed to drop. Cerebras claims that its response time for large models is 15 times faster than GPU-based systems. This is crucial for OpenAI, as faster, real-time responses allow users to complete more tasks, spend more time on the platform, and perform higher-value workloads.
Sam Altman was both a client and a financier.
The relationship behind this deal is quite deep, as OpenAI CEO Sam Altman is a personal investor in Cerebras. The collaboration between the two companies had been brewing for some time; as early as 2017, shortly after their founding, the teams frequently met to share research findings and discuss the possibility of further cooperation.
Negotiations for this collaboration began last fall, and a letter of intent was signed before Thanksgiving. Cerebras CEO Andrew Feldman stated that the key driver of the market is the extraordinary demand for "fast computing."
Cerebras' valuation doubles; OpenAI seeks "non-NVIDIA" solutions.
Encouraged by this large order, Cerebras is in talks for a $1 billion funding round, which could potentially boost its market valuation to $22 billion, almost three times its original value. The company previously filed for an IPO in 2024 but subsequently withdrew it; however, it still plans to proceed with its listing.
For OpenAI, this is also part of its strategy to diversify its computing power supply chain. In its search for cheaper and more efficient alternatives to NVIDIA, OpenAI had previously announced a partnership with Broadcom. Collaborative development of customized chipsAnd signed an agreement to use AMD's MI450 chip.Accelerate its computing system.
